DataArt on the Leading Edge of Food Recognition

DataArt on the Leading Edge of Food Recognition

The DataArt Orange initiative spent lots of development efforts for its food recognition R&D project. And finally it feels that the market is ready for the technology. Last week Google announced at the Rework Deep Learning Summit an artificial intelligence project to calculate the calories in pictures of food you have taken. According to The Guardian, “the prospective tool called Im2Calories, aims to identify food snapped and work out the calorie content”. There is not much information about the project and what algorithms are available at the moment, but what is available indicates that Im2Calories will utilize a similar approach used by DataArt’s Computer Vision Competence Centre researchers in their Eat’n’Click project.
Read More »

DataArt Orange to integrate its food recognition engine with Apple HealthKit and Voice Recognition

Eat’n’Click is an application developed by DataArt that helps people track the nutritional content of foods they consume. The app automatically tracks calorie information by scanning photographs of food.

Its prototype is available on the App Store and can recognize a number of fruits with more than an 85% success rate. Recently the application was demonstrated by the DataArt team during the Health 2.0 Europe 2014 event in London and was highly appreciated by its participants.

Following industry trends and the high attention Apple HealthKit is receiving, DataArt’s Research Lab decided to move forward by automating the tracking of users’ nutrition habits. DataArt is planning to build a PoC prototype to integrate with Apple HealthKit. Also, we plan to implement a voice recognition engine with it. The aim of DataArt’s Orange R&D initiative is to reduce the gap between humanity and computers. And we strongly believe that this step will be highly effective.

FoodR Slim to be integrated with Apple HealthKit and Voice Recognition
Read More »

Fruit recognition – some improvement

The quality of visual fruit recognition PoC was improved by adding texture and RGB color space features information to the feature vector. Same as in our previous experiment, we took the neural network as a classifier. The network learns with error of 0.155. The results are given below: fruit recognition In general, the quality of recognition was increased in 0.112%.
Read More »

Fruit Recognition – Continuing Research

Fruit Recognition – Continuing Research

DataArt Research Lab continues to publish the results of fruit recognition using the Color Distance method. The Color Distance method showed good result as an engine for fruit recognition. Just for testing purposes, we took 15 classes of fruits (10 ‘ideal’ samples for each class) and a simple classifier of Euclidian distance. But such an approach has serious disadvantages:
  • in everyday life we do not deal with ideal pictures: the photos may contain different distortions, irregular brightness  etc.;
  • the classifier of Euclidian distance cannot provide us with real time work, particularly if we have a lot of classes and etalons.
So our next step is to use a more complicated real time working classifier and increase the sample set with new ‘real’ images.
Read More »

DataArt Announces ORANGE – the Missing Piece in Nutrition Tracking

New R&D initiative provides food recognition technology to calculate calorie intake NEW YORK – March 27, 2014DataArt, a leading custom software development company that builds advanced solutions for select industries, today announced the first results of DataArt ORANGE, a series of research and development projects that aim to automate the tracking of users’ nutrition habits. The DataArt ORANGE program automatically tracks calorie information by scanning photographs of food. DataArt ORANGE technology can currently recognize over 100 foods with an 85% success rate.
Read More »

DataArt Research Lab publishes Color distance method testing results

DataArt Research Lab publishes Color distance method testing results

The color distance method has been tested using 15 fruit types, each type having 10 photos. The following recognition score calculation algorithm has been used:

  • Three guesses (ordered best to worst) are considered.
  • A correct guess placed first obtains 10 points, placed second – 5, placed third has 1.
  • For all tests for a particular fruit type the obtained points are summed up and divided by the absolute maximum result 10 * N, where N – the number of tests for the class, thus yielding the classification quality for the type.
  • The overall classification quality for the whole fruit set, the individual classification qualities for every fruit type are averaged.
Read More »

DataArt Research Lab to experiment on finding and proofing feature extraction methods suitable for food recognition tasks

img_area_phantom
Event Image
Meals named the same rarely look similar. This is not only due to different people cook differently – in the computer vision sense, meals are combinations of areas (spots) with different color, texture, shape each. This makes typical image recognition principles less suitable for food image recognition, as we cannot rely on either form or relative position of the image parts. Typically, if local peculiarities of objects being detected cannot be caught, integration feature extraction methods take over differential one – e.g. in our current food image classification engine we mostly rely on combined histogram and texture parameters for the whole image. This approach shows relatively good results unless the meal we’re trying to classify appears to have no noticeable texture features.
Read More »